Modelling studies may correctly predict COVID-19 cases only two weeks into future: scientists
New Delhi, Jul 2 () Mathematical models predicting the trajectory of COVID-19 cases are only approximate guides to the truth and can, at best, be used for projections up to two weeks, say scientists, tempering excitement over the many 'forecasts' of the disease.The note of caution comes as various modelling studies have used mathematical tools to predict different scenarios – one, for instance, finding that the cases in India will peak in November and another saying in March that the country would have 13 lakh cases by May. Advertisement
However, these studies are to be treated with utmost scepticism given the many variables involved and the fact that knowledge on COVID-19 is still evolving, scientists said.
"I think, if you want to use models for correctly predicting cases, two weeks into the future is the best one can do, and if one's lucky the trend will continue for longer, but you can't bank on that," Sitabhra Sinha, professor of Computational Biology and Theoretical Physics at Chennai's Institute of Mathematical Sciences (IMSc), told .He added that he is sceptical about specific long-term predictions such as "the peak will occur on so-and-so month".
An editorial in ICMR's Indian Journal of Medical Research (IJMR) earlier this month said mathematical models on severity of the COVID-19 pandemic in India carried a "strong element of bias" and used assumptions to predict cases and deaths which proved to be far from real.The editorial followed an unpublished study that found the peak stage of COVID-19 in India had been delayed by the eight-week lockdown and may now arrive around mid-November when a paucity of isolation and ICU beds and ventilators can arise. Sinha was unconvinced about the veracity of the claim. Advertisement
He told there are too many uncontrollable variables operating in modelling studies to say with any degree of accuracy about what's going to happen in November or for that matter in August.
He also referred to a March prediction by an international team of scientists, including from the Johns Hopkins University, that India could face between 1 lakh to 13 lakh confirmed cases of the novel coronavirus by mid-May provided the trend in the growing number of COVID-19 cases continued."However, it didn't come to pass fortunately -- because of the lockdown, which is something that the modellers of the study could never have foreseen," Sinha explained. Advertisement
"Given that the growth in the number of cases depends so sensitively on containment measures that are put in place from time to time, which we can't know beforehand, how are we going to make accurate long-term projections?" he asked.
He added that even with knowledge of the containment measures, there is little information about the level of compliance of physical distancing and public hygiene norms in different places."Without exact knowledge of these social variables that shape the epidemic trajectory, I think it's unfair to expect that modellers would be able to foretell the future," said Sinha. Advertisement
"This is true no matter how many bells and whistles you add to the model - in fact the more complicated a model, the more parameters you need to fit from empirical data which is noisy and error-prone. So at the end of the day, the fancier models may not be much of an improvement on the simpler models," he said.
Gautam I. Menon, professor, Departments of Physics and Biology at Ashoka University in Haryana endorsed his scepticism.Models, he said, are only "approximate guides to the truth" and must be constantly benchmarked against data. Advertisement
"Our knowledge of COVID-19 itself has been evolving. For example, the importance of asymptomatic carriers has only come to be appreciated more recently. In addition, data useful for models are not transparently available," Menon told .He said one should also look into the quality of the data that goes into these models. "All models make approximations and also take in data that is often imperfect. The question is, do they give the right intuition about the progress of the disease and reasonable quantitative predictions regarding, say, the requirements for hospital beds etc. within these limitations," he said. Advertisement
"Given these caveats, I would say that models provide information to inform policy that is simply irreplaceable. The quality of models is improving, both as we know more about the disease as well as improve the quality of inputs to the model," Menon added.
Adding his voice to the debate, Dibyendu Nandi, professor of physics at Kolkata's Indian Institute of Science Education and Research (IISER), noted that the classic approach seeks to model the progression of the disease based on rates of movement of susceptible individuals exposed to the infected, recovered or deceased people.The basic premise of these models, he said, is based on the theory of epidemiology that has been tested against past pandemics, and is proven to be sound. Advertisement
However, he added, model behaviour and in particular model predictions are strongly dependent on some parameters that are not very well known in the initial phase of a disease.
"So instead of making exact predictions, in my opinion these models are more suitable for assessing various possible future scenarios," Nandi told ."For example, how would the disease progress if the containment measures are less efficient than planned? What would happen if we remove all containment measures, including testing and quarantining?" Advertisement
Nandi said his experience with physical models indicates that qualitative trends coming out of these models must be stressed more than exact numbers.
Sinha added that the true utility of modelling is to test out various policies and ideas for controlling the epidemic in silico -- or perform them via computer simulation -- before you try them out in real life."This is important because a wrong step in the latter can end up costing many lives," he said. Advertisement
"You'd prefer if you can get the answer from modelling rather than finding out through trial and error in real-life," Sinha said.Nandi reiterated a caveat about the application of mathematical models to infectious disease progression. "The dynamics of pandemics depend on human social behaviour and the efficiency of containment measures that are impossible to predict. This possibly explains why exact predictions do not work well," he said. Advertisement
However, that does not take away from the fact that these models are extremely useful in qualitatively assessing various plausible scenarios, he said. Well constrained models may also generate meaningful short-term forecasts, he added.India's COVID-19 tally on Thursday crossed the six lakh mark while the death toll rose to 17,834, according to the Union health ministry data. SAR MIN MIN MIN
(This story has not been edited by Business Insider and is auto-generated from a syndicated feed we subscribe to.)